The Ceiling Became the Goal: Google Ads' August 17 Change and the End of Lucky ROAS

On Monday, August 17, Google changed how Smart Bidding works in accounts worldwide. There was no keynote, no flashing banner in the interface, no email with an all-caps subject line. If you manage budget-constrained campaigns on Target CPA or Target ROAS, the change has been live in your account for four days — and most of the market still hasn't opened the spreadsheet to look.

The official description has an almost comic innocence to it: budget-limited campaigns using a target-based bid strategy will now perform "more consistently toward your bid target." Read that again, paying attention to the direction. Toward the target. Not below it.

Translated into the language of whoever pays the invoice: if your Target CPA was $50 and the campaign had been delivering $35, Google is going to stop delivering you $35.

Nearly everything written about this was published before the 17th, in the "what to fix before the deadline" format. The deadline passed. This article is the other half of the conversation: it already happened, it's in your historical data, and the interesting thing now isn't the preventive checklist — it's what the next thirty days will reveal about your operation.

What changed, without the euphemism

The facts, straight from the Google Ads documentation:

Starting August 17, 2026, budget-limited campaigns using a target-based bid strategy will consistently optimize toward the stated target, including when you adjust budgets.
It affects Search, Shopping, Performance Max, Demand Gen, Display, Hotel, and Travel.
It does not affect App campaigns, Video reach, or Video view campaigns, which keep the old behavior.
It applies across Google Ads, Search Ads 360, Display & Video 360, Google Ads Editor, and the API.
Campaigns that are not budget-constrained do not change behavior at all.

That last point is the most important and the most ignored. The trigger for this change is the "Limited by budget" status — the label plenty of people display in meetings like a trophy, proof they found a channel that can absorb more money. That trophy just became a bill.

Google also opened a bid target adjustment tool inside the platform on July 6, reachable from account notifications or the Campaigns page. In other words: there were six weeks of warning. If you didn't use it, it wasn't because anyone hid it.

Google isn't lying. Neither are the practitioners.

Here's the part most coverage blurred, and it's worth untangling carefully, because the confusion is leading good people to the wrong conclusion.

When the market started reading the change as "Google unlocked the system to spend more money," Ginny Marvin, Google's Ads Liaison, was blunt: that claim is inaccurate, and the change will not lead to spend increases. Google reinforced two points alongside it — daily and monthly budget limits are always respected, and the company will not automatically adjust your budgets or your campaign targets. If you wanted to preserve your prior performance, the manual adjustment was on you.

At the same time, experienced practitioners pushed back with legitimate irritation. Joey Bidner pointed out that several of his best-performing accounts intentionally run with low tROAS or high tCPA targets, using that slack as a strategy. Nils Rooijmans warned that anyone leaving targets untouched would see "decreasing efficiency of spend." And Xavier Mantica asked the sharpest mechanical question of all: how exactly will the algorithm hit the target — will it just increase CPCs until it inflates my CPA?

Both sides are right, and that's why the argument stalled. They're talking about two different variables:

Total spend does not go up. The budget is a hard ceiling. Google is technically correct, no asterisk needed.
Cost per outcome does go up. Same money going in, more aggressive bids, auctions the system used to skip. Same spend, higher CPA, fewer conversions at month's end.

There's no contradiction between those two sentences. There's only a choice about which one goes on the slide. Google measures spend. You're measured on outcomes. The change moves precisely the variable that isn't in the announcement.

And to answer Mantica: yes, that's the mechanism. With the budget capped, the only route to a higher CPA is buying more expensive inventory — entering auctions the system's conservatism had been avoiding. There is no other lever.

The target was never a goal. It was a ceiling.

Now the part that actually matters, because it isn't really about Google Ads.

For years, the "Target CPA" field did two jobs at once, and almost nobody separated them: it was a goal for the algorithm to chase and a tolerance limit you declared you'd accept. In campaigns with budget to spare, it behaved like a goal. In budget-constrained campaigns, it behaved like a ceiling — the system had more qualified demand than money, picked off the cheapest opportunities first, spent everything, and delivered well below what you had authorized.

The gap between what you authorized and what you paid has a name in economics: consumer surplus. You said "I'll pay up to $50" and you paid $35. That $15 was yours, and it was structural — it didn't come from your talent, it came from the system's caution in the face of a tight budget.

What happened on the 17th was the collection of that surplus. Google didn't steal anything and didn't hide anything. It simply stopped charging you less than you had said you were willing to pay.

Notice the incentive design, which is elegant and requires no conspiracy theory to work. With the slack eliminated, your measured efficiency now equals your declared tolerance. And if you want more conversions, exactly one path remains — the same one Google's documentation recommends out loud: raise the budget. The change doesn't move money out of your account today. It converts an invisible efficiency gain into an explicit budget decision. That's far better incentive engineering than a price increase, because it survives any audit: you received exactly what you asked for.

The question nobody wants to ask in the meeting

I've written here before that your ROAS is a lie. This is the cleanest case for that thesis to appear in years, and it arrived free, in the form of a natural experiment Google just ran inside your account without asking permission.

The question is this: how much of your performance case was skill, and how much was system slack?

If you had a budget-constrained campaign with a CPA well below target, and you presented that as the result of management — creative optimization, audience refinement, negative keywords, campaign structure — the next thirty days will tell you how much of that was true. If the CPA climbs until it touches the target while nothing in your operation changed, the answer is uncomfortable: part of your performance was an artifact of the algorithm's timidity against a short budget.

That's not shameful, and it isn't an accusation. It's information that was previously impossible to obtain, because there was no way to separate the two effects. Now there is. An honest operator will use this window to calibrate what they actually control. A less honest one will spend September explaining that "the market got more competitive."

There's a message for the other side of the table too. If you're the client or the business owner and your CPA jumped 30% in September with no change in strategy, before you switch agencies: ask whether those campaigns were budget-constrained and what the gap between target and reality looked like in July. Your agency may have done nothing wrong — or it may have skipped an adjustment it had six weeks of warning about. Both hypotheses are verifiable in five minutes of reporting.

How to read your account this week

We're on day four. That changes what's worth doing, and the order matters.

Don't look at Performance Planner right now. Between August 17 and 31 it's working from a behavioral model that just shifted underneath it. Any projection it gives you in this window is fiction. It becomes useful again in September.

Don't stack fixes. The pure observation window of the first few days has passed, but the principle holds: Smart Bidding reacts to target changes in real time. If you touch targets, budgets, and creative in the same week the entire system changed, you forfeit any chance of knowing what caused what. One variable at a time.

Pull the right list. Export every campaign that showed "Limited by budget" in the last twelve months — not just the ones carrying the label today. Campaigns that were intermittently constrained count, and they're the easiest to miss. Filter for Target CPA, Target ROAS, and Target CPC (the last one only in Demand Gen).

Prioritize by money, not by gap. This is the mistake that will cost people a week of sorting the wrong column. What matters is gap multiplied by monthly spend, not the size of the gap. A 2x gap on a $400/month campaign is noise. A 20% gap on an $80,000/month campaign is your quarter. Sort by financial exposure and work top down.

Use the right data window. Thirty days for high-volume campaigns; sixty to ninety for thin-data campaigns. And judge by your business's conversion cycle, not the calendar — if your lead takes eighteen days to close, reading results on day seven isn't analysis, it's anxiety.

The five exits

For every campaign on that list, there are five possible decisions. None of them is "wait and see."

Keep the target as is — but only if you can articulate why, out loud, without using the word "historically." Deliberate slack is a legitimate strategy; forgotten slack is not. If you keep it, give it budget buffer and consider consolidating campaigns that share a target under portfolio bidding.
Match recent actuals — Google's default recommendation. Delivering $35 against a $50 target? Set it to $35. It's the safest exit and the one that best preserves the result you already had.
Recalculate from unit economics — the right path when the target was inherited and never stress-tested. What can you actually pay per conversion given your ticket size, your margin, and your close rate? Model it in the bid simulator before applying.
Move in steps — nudge slowly rather than jumping at once. A large adjustment throws the campaign into a learning phase, and that volatility will mask the real effect of the change.
Raise the budget instead of the target — valid when the target is genuinely correct and demand exists above your cap. After the 17th, budget increases scale at the stated target, which is the one part of this change working in your favor: results become predictable as you scale.

Two things that are not solutions, despite being suggested around the internet: data exclusions and manual bid limits. Both carry performance-fluctuation risk and neither addresses the cause.

And the most important part: this isn't an August fire drill. Target versus actual on budget-constrained campaigns doesn't stay settled just because you fixed it once. Performance moves, the gap reopens. This is a monthly routine item now.

The lesson that outlives Google Ads

If you forget everything else, keep this sentence: every field a platform lets you fill with a tolerance number is a price you agreed to pay.

You didn't type that $50 Target CPA thinking "I want to pay $50." You typed it thinking "I won't pay more than $50." The platform read the first version. And as long as the difference between those two readings was landing in your pocket, nobody needed to notice the ambiguity.

That same ambiguity is scattered across your entire stack. The minimum ROAS you configured in bidding. The maximum CPL you declared in your media tool. The discount you authorized sales to give "if they need to." The delivery date you promised with padding built in. Every one of those numbers is a tolerance that some system — or some person — may one day decide to consume in full, without breaking a single rule and without telling you.

The practical conclusion isn't paranoia. It's to audit, once a quarter, every tolerance number you've declared to an automated system, and answer two questions: how much slack exists between this and reality today? and if that slack vanished tomorrow, would I survive?

Slack you don't measure is slack you're going to lose. Google just proved it on a Monday, with no notice in your account.

If you want someone to go through that campaign list with you before September closes, this is exactly the kind of work I do. Get in touch.

References

— views— likes